Estimation of Vehicle Roll Angle

Haldex Traction can with their Active Yaw

Control prevent unwanted handling when
driving a vehicle, by applying an extra
torque with their all wheel drive system.
To be able to calculate when and how
this torque should be applied, it is
important to the state information of the
vehicle. When driving on banked roads,
vehicle
dynamics
and
sensor
measurements are changed compared to
when driving on a flat surface. Because of
this it is desirable to know the road
banking or vehicle roll angle.
The vehicle roll angle was estimated using
sensors already existing in modern
production cars; lateral acceleration ,
longitudinal speed
, yaw rate
and
front wheel angle . The acceleration and
yaw rate sensors are positioned in the
vehicles center of gravity. This is also the
origin of the right hand based coordinate
system used for defining sign convention.
Thus the x-axis points forward, the y-axis
to the left and the z-axis upwards.
The acceleration measured by the
acceleration sensor occurs due to lateral
forces on the tires when cornering and
due to gravity affecting the sensor when
the vehicle rolls around the x-axis.
Yaw rate is the angular velocity the vehicle
has around the z-axis. Thus the yaw rate
sensor will measure how fast the vehicle is
turning.

Theoretical solution
The lateral accelerometer is, when tilted
around the x-axis (travel direction),
influenced by the normal forces acting on
the vehicle due to gravity. By knowing
how much the roll angle can be calculated.
Unfortunately
there
are
more

accelerations acting at the accelerometer;

centripetal force and lateral speed
derivative . The centripetal force occurs
when an object follows a curved path. The
force is always directed towards the
center of curvature of the path. Lateral
speed are present due to lateral slip when
turning, it is often very small, but will
influence the accelerometer greatly when
making sudden turns. The equation for the
angle calculation becomes:
(1)
It is derived from the accelerations acting
on the vehicle driving in a banked curve,
as seen in Fig 1. It is important to separate
the environmental coordinate system
denoted with an apostrophe, from the
vehicle coordinate system.

Figure 1 - Accelerations acting on vehicle in a banked

curve

When testing the equation in a vehicle

simulator at Haldex called VehSim, eq. (1)
performed as seen in Fig. 2. The test
simulation vehicle was in this test driving
in a 180 degree turn with a road banking
of 20 degrees with start and finish at 0
degrees.

Figure 4 Lateral forces with respect to slip angle

Figure 2 - Vehicle chassis roll angle

All units in eq. (1) are measured with

sensors in the car. The problem is to
obtain a good estimation of . Without
this parameter the roll angle estimation
will deviate several degrees when sudden
turns are made.

Vehicle model
By using a simple bicycle model,
representing a vehicle, the lateral
acceleration can be estimated. Figure 3
shows an overview of the forces and
velocities acting on a vehicle represented
as a two-wheel bicycle model.
When a vehicle turns, slip angles will be
present due to lateral velocity. The slip
angles
can
be
calculated
using
longitudinal-, lateral-velocity, yaw rate,
length to the center of gravity and front
wheel angle. If having the slip angles,
lateral tire forces can be calculated using a
tire formula that gives the relationship
between slip angle and lateral tire force.
There are many different tire models with
different properties. Fig. 4 shows results of
the lateral force obtained with the three
most common models; linear, exponential
and magic formula.

The magic formula gives the results that

are closest to the real forces if they are
measured. The downside with the magic
formula is that it is very complex and
computationally heavy. While the slip
angle is small, all models perform similar.
When the slip angle grows the linear
model performs poorly while the
exponential
model
still
achieves
acceptable results.
The friction coefficient () is another key
parameter used to estimate the lateral
forces. This parameter gives the friction
relationship between the tire and the
surface that the tire is on. In Fig. 5 the
relation between the friction coefficient
and lateral tire force is show.

Figure 5 Lateral force with respect to friction

Figure 3 - Bicycle model

Based on the estimated lateral forces the

dynamics of the vehicle can be
determined.

Observer
The model is unfortunately not perfect
since it is a simplified version of a vehicle.
This will cause the estimated dynamics to
differ from the real dynamics. The results
from the model can be improved by
adding an observer. The observer will
make the model converge towards the
real dynamics by correcting it.
The correction is based on the predicted
parameter value from the model
compared with the measured value like
the Fig. 6 shows.

the model estimates poorly or that the

vehicle skids in a way that affects the yaw
rate sensor but not the lateral acceleration
sensor. These errors in the calculated
angle must be detected and actions must
be taken.

Figure 7 Calculated and measured roll angle in

extreme driving situation

Figure 6 Basic function of an observer

There are different observers with

different properties and complexity. One
often used observer is the Kalman filter,
which can be modified to an Extended
Kalman Filter to handle nonlinear systems.
The two observers tested in this thesis
were the powerful and complex Extended
Kalman Filter and the much simpler
Average Observer. The result in this case
was that the Average Observer performed
almost as good as the more complex
Extended Kalman Filter. The Average
Observer is much lighter on computations
and easier to tune and therefore it was
the final choice of observer.

Fig. 7 clearly shows that large peaks in the

calculated angle deviate from the
measured angle. Methods to identify
when the calculated angle is wrong were
developed and implemented. When errors
occur the calculated angle needs to be
filtered. This filter was also developed and
implemented and the result is shown in
Fig. 8.

Figure 8 Filtered calculated angle

Error detection

Results

In some cases the angle calculation

algorithm calculates an incorrect angle.
This is due to different problems such as

One thing that is important to take into

consideration is that the vehicle is affected

both by the banking angle of the road as

well as roll due to its suspension.

the results. There are still much work that

needs to be done with error detection and
what to do when an error occur.

While the vehicle is driven under normal

circumstances the results of the angle
calculation are acceptable. Fig. 9 shows
the results in VehSim when driving in a
curve with 15 degree banking.

Figure 9 Results in VehSim

Fig. 10 shows the results from a real test

drive at Ljungbyhed. The vehicle was
driven in a zigzag pattern.

Figure 10 Results from real test drive

The final conclusion is that the angle

calculation algorithm was greatly
improved from Haldexs own version. The
algorithm produces good results under
normal driving conditions. When the
vehicle is driven hard and large slip occur

Lund, January 11
Herman Lingefelt and Anthon Nilsson